Mathematical problem-solving skills are a critical 21st-century competency, yet their development still faces various challenges in the digital age. This study aims to explore teachers’ perspectives on students’ mathematical problem-solving skills, analyze teaching practices through the Theory of Didactical Situations framework, and formulate a reconceptualization of mathematical problem-solving skills relevant to the era of artificial intelligence. The study employed a qualitative approach with an exploratory design involving 11 junior high school mathematics teachers selected through purposive sampling. Data were collected via semi-structured interviews and analyzed using the Bogdan and Biklen model. Findings indicate that mathematics instruction remains dominated by a procedural approach that reinforces students’ reliance on examples and teacher authority. Students struggle to understand non-routine problems, design independent strategies, engage in reflection, and verify solutions. Furthermore, the use of digital technology remains instrumental and does not yet support exploration or critical thinking. From a TDS perspective, the digital milieu has not yet functioned as an exploratory environment, while the didactic contract still positions the teacher as the primary authority. Based on a synthesis of empirical findings and studies on AI hallucination and Retrieval-Augmented Generation, this research proposes an expansion of the problem-solving framework by adding a dimension of digital information verification and formulating a conceptual model of the Didactic RAG-System to support mathematics learning in the era of artificial intelligence.